No One Knows
Fun anecdote, watching the AI Doc, how I became an Apocoloprimist. (Which btw really is in the samw spirt of film titles for films as Dr. Strangelove or: How I Learned to Stop Worrying and Love the Bomb (1964), directed by Stanley Kubrick and perhaps To Wong Foo, Thanks for Everything! Julie Newmar ).
The Director slash interviewer asks everyone what will happen. There is the Existential Risk tribe and the Accelerationist tribe. Also the movie is sponsored by Tristan Harris who is bipartisan here. (Raskin as well).
But Daniel also speaks to several CEOs , Sam Altman , Dario Amodei, Demis Hassabis. And they know not what will happen. All they can say that is meaningful is that AGI is an auction that goes to the winner who takes the most safety shortcuts.
No one can predict much of anything anyway
But anyway, to me this is as expected, though many people (only some people maybe?), do put such high valuations on the frontier labs because they promise to essentially do what it seems no one can really do: make promises.
AI models are sold –indirectly?–as oracles, as genies (dqtabricks assistant ), as fortune tellers , even though interestingly they are apart from “predictive models” of now classic ML.
I like anecdotes like this as reminders that, in the vast majority of the important business questions no one knows the answer. Because they are not statistically answerable questions. There is not enough data. All that can be done is to –cherry?– pick the data that favors your case. Or you can run simulations, of course , with your assumptions.
Frontier lab models are not inherently offering anything new here but yes lots of (some?) people want to believe they will magically derisk their problems.
But funny thing is that life still needs to be lived forward and there are no shortcuts to seeing tbrough the fog of war.
references
- Kahneman, Tversky, on planning fallacy